For any team or business

Practical AI workflows, without the theater.

We publish what we learn about practical AI workflows: the articles and First Look, where you can watch the checks. We will publish the studies we run on our own site.

Read the articles See the studies Open First Look

Example workflow

A weekly report, with the sources attached.

An example of the kind of workflow the articles describe. Not a client's result.

Inputs

Last week's notes
The open ticket list
The source brief

Steps

01 Read the notes, tickets, and brief
02 Draft the report from those sources only
03 Attach a citation to each claim
04 A person checks it before it goes out

Output

Three items need a look: the brief's date (source brief), two tickets still open (ticket list), and one note that is not on the ticket list (last week's notes).

Cited from the source brief, the ticket list, and last week's notes.

Writing

The latest writing

Notes on building and using AI workflows.

All articles

Article15 Oct 2026

Why every AI Overviews study seems to say something different

AI search studies keep contradicting each other. Why that happens, what four 2026 studies actually measured, and how to read the next headline number.

Read the article

Article15 Oct 2026

How do you get cited in AI answers, not just ranked?

Ranking well on Google does not mean AI answers will cite you. Four 2026 studies and our trace data show the gap between being found and being credited.

Read the article

Article12 Oct 2026

AI agents can favor some sources over a better fit

A new study suggests the site an item comes from can sway an AI agent's choice, even when another item fits the request better. Give agents fewer gaps to fill with assumptions.

Read the article

Studies

What we study on our own site

No study is published yet. The decision-model studies are next, and each one will state the question, the method, the sample, the dates, the findings, and the limits. First Look is the live checks page.

Read the studies Open First Look

The method

Start with the work, not the model.

Pick the job first: research, analysis, reporting, planning, review. Then decide what context, tools, checks, and output it actually needs. A workflow written this way has the same five parts.

Read the method

Source materialFiles, notes, sites, transcripts, spreadsheets, and records that ground the task.
Tools and agentsSearch, APIs, databases, MCP servers, and browser actions, with clear boundaries.
Reusable instructionsPrompts, skills, templates, and checklists that turn one-off requests into workflows.
Evidence and reviewCitations, checks, limits, and human judgment before anything is trusted or sent.
Useful artifactsReports, briefs, pages, plans, and dashboards people can inspect and reuse.

WebMCP, running here

Try the tools on this site.

Most sites make agents scrape and guess. Here, the forms declare themselves as tools an agent can call, with named inputs, plain descriptions, and a person reviewing anything that gets sent.

11tools a browser agent can call on this site

Declared, not scrapedTools are declared in the page markup, so an agent calls the form instead of guessing at the layout.
Bounded on purposeEvery tool here is read-only or reviewed intake. Nothing is answered or acted on automatically.
Same result for peopleWithout an agent, the same buttons run the same functions. Accessible labels serve both.

Checks the title, meta, H1, declared tools, llms.txt, and robots.txt on one page. Open First Look to watch an agent use it.

index.htmldeclarative WebMCP

<!-- The Ask form below, as an agent sees it -->
<form action="/api/ask" method="post"
  toolname="ask_rafalai"
  tooldescription="Submit a practical AI or
    educational topic question to RafalAI for review.">

  <input name="email" type="email" required
    toolparamdescription="The sender's email
      address for the response.">
  <!-- name, topic, message … -->
</form>

Tools on /first-look

Explain this pageexplain_this_page
List the public pageslist_public_work
List the briefslist_documents
Explain the briefexplain_document
Show the briefshow_document
Set a URLset_their_url
Run a first lookrun_first_look
Check pasted HTMLscan_html
Read the current lookget_first_look
List questionslist_questions

Work with us

We are a new practice. These three pages are how the work is done. They are not a record of client jobs. Each one ends with a way to write to us.

About

What we do

RafalAI is a new practice. We write about practical AI workflows and show the checks in First Look. We will publish the studies we run on our own site. We do not have past client work to show yet.

Write to us on the contact page, or follow along on LinkedIn.

Ask a question

Stuck on something? Ask.

Questions about using AI for real work: a workflow that isn't holding up, a prompt, a tool, a source you can't verify, or a claim you want checked.

A person reads every question.
Your browser agent can fill this form for you. It is the ask_rafalai tool.
Want something built instead? Tell us about the work.